Global Capital Allocation

Foreign assets and liabilities can be decomposed into four major categories.² The first category is Portfolio Investment and it covers holdings of equity and bond securities, excluding those assets held as reserves (see below). In addition, shares in investment funds such as mutual funds and exchange-traded funds are considered part of equity holdings irrespective of the underlying investment focus of the funds. The second category is Foreign Direct Investment (FDI), which identifies both equity and debt investments where the investor holds a significant degree of control over the recipient of the investment, as well as those that give rise to such control or influence. Generally, the threshold for control is defined to be when an investor holds 10% of the outstanding equity securities of an enterprise. FDI debt consists of lending between parties with a control or influence relationship. The third category is the poorly named Other Investment. This category consists mostly of the cross-border activity of the banking sector such as bank loans and deposits, as well as miscellaneous categories like SDR allocations and other equity. The fourth category is Reserves. These assets (generally bonds) held mostly by central banks are not included in portfolio investment on the asset side of the holding country. However, these same holdings will be accounted for as portfolio liabilities of the country issuing the corresponding securities (there is no Reserves category in foreign liabilities).

The composition of a country’s balance sheet can be a key channel through which international shocks are either amplified or dampened. Consider, for instance, a net debtor country whose liabilities are composed entirely of portfolio equity and debt. These types of liabilities are often referred to as being driven by “hot money,” a form of foreign capital particularly flighty during crises. Generally, policymakers worry that a country may be more vulnerable to capital outflows in response to global shocks if foreigners find it easy to sell their assets (Forbes and Warnock (2012); Caballero and Simsek (2020)). By contrast, if a country’s liabilities are primarily in the “Other” category, then the country is particularly reliant on foreign bank finance, potentially linking its financing conditions to the health of the global banking system.

After experiencing a series of sudden stops, policymakers in emerging economies have aspired to attract what they perceive to be a more stable form of foreign financing: foreign direct investment (FDI). Since FDI involves a long-term commitment and presumably an institutional commitment on the investor side, the idea is that it would be harder for investors to liquidate the investment in bad times. While this is true for traditional FDI, such as when a foreign firm builds a plant in the country, new research has shown that in recent years FDI often consists of intracompany transfers for tax purposes (Damgaard, Elkjaer, and Johannesen (2019); Tørsløv, Wier, and Zucman (2023)), or masks portfolio investment due to securities issuance by foreign affiliates of domestic firms (Coppola et al. (2021)).

In Figure 3, we explore the changing nature of cross-border investment patterns over the last several decades. We document a major decline in the share of assets and liabilities in the Other Investment category. An overwhelming share of cross-border investment in the 1970s, 80s, and 90s was intermediated by the banking sector making the Other category very large. The fall in cross-border bank intermediation is offset by the rise of market-based finance, with the share of cross-border assets in portfolio investment rising rapidly from 1980 to present. In addition, we observe an increase in the share coming from FDI. Finally, at a global level, we observe a decline in the share of assets accounted for by reserves.

In Figure 4, we split the sample into emerging and developed markets. ³ The fall in global reserves as a share of cross-border assets is driven by the effective disappearance of reserve holdings by developed economies.⁴ The liability side of emerging market balance sheets displays a striking change during this period. While in the mid-1990s, 62% of these countries’ liabilities came in the form of bank-dominated “Other” investment, this share was only 26% by 2021. For developed economies, while the decline of Other finance is also visible, it is the rise of cross-border portfolio investment on both the asset and liability side that dominates the historical trend.⁵

While we can learn much from the composition of a country’s assets and liabilities at the multilateral level, these patterns leave many questions unanswered. The more recent literature has made a remarkable degree of progress in characterizing the geography of investment around the world using a new generation of bilateral data sources. In particular, the IMF Coordinated Portfolio Investment Survey, introduced in 2001, was a significant advance for research on global capital allocation. The IMF introduced a survey of countries’ bilateral portfolio positions, with an increasing emphasis in recent years on the sector and currency of investment. This allowed researchers to go beyond asking only whether countries are home-biased, and instead explore where countries allocate their international investment among the many foreign destinations. An early literature building on these new data adapted the gravity regressions from goods trade (Anderson and Van Wincoop (2003); Head and Mayer (2014)) to trade in financial assets. Portes and Rey (2005) characterize the bilateral pattern of portfolio investment in much the same way trade papers characterize bilateral goods trade and early micro-foundations of the gravity literature for financial flows and positions can be found in Martin and Rey (2000) and Martin and Rey (2004). Okawa and Van Wincoop (2012) provides an insightful formulation of the theoretical underpinnings of gravity models in international finance.⁶

The IMF Coordinated Direct Investment Survey (CDIS) plays a similar role to CPIS but focuses on bilateral FDI investments instead, with countries reporting their outward and inward FDI separately. For bank loans, researchers frequently use the Bank for International Settlements Locational Banking Statistics (BIS LBS). In addition to geography, the BIS LBS statistics data includes information such as the sector of the investor and issuer and the currency of the loans. Reserves are substantially harder to decompose either bilaterally or by currency.⁷ Didier, Schmukler, and Peter (2023) add up the data from these bilateral datasets and find an increase in bilateral capital allocations of emerging economies to other emerging economies.

The analysis of the geography of capital allocation naturally brings up the question of what defines the country of destination of an investment and how it is measured in these public data. Financial globalization and the rise of multinational firms has made researchers’ life more difficult but also more interesting in this respect.

One of the most promising developments in the last few years has been the rise of theoretical frameworks that seek to explain the network of bilateral investment positions and use the model structure to identify and quantitatively assess the key frictions needed to match the data. By leveraging the full network of bilateral global capital allocations rather than simply analyzing multilateral external positions, this research agenda opens the door to a better analysis of the nature of market segmentation.

Koijen and Yogo (2019) apply a demand-based asset pricing system to bilateral data on portfolio investment positions to decompose the drivers of exchange rates and asset price movements around the world. Jiang, Richmond, and Zhang (2022)a build on this demand-based framework to quantify the drivers of the dynamics of the US net foreign asset position, and in subsequent work apply their framework to decompose dollar exchange rate movements (Jiang, Richmond, and Zhang (2022)b). Shen and Zhang (2022) adapt this framework to study cross-border bank lending and document how differentiated credit supply elasticities affect cross-border risk-sharing.

Pellegrino, Spolaore, and Wacziarg (2022) characterize the network of global cross-border investment in a multi-country general equilibrium framework. They use their estimated model to argue that barriers to global capital allocation account for sizable welfare losses. Combining an Armington model of trade with an Eaton and Kortum (2002) model of financial investment, Kleinman et al. (2022) jointly analyze the network of trade and capital allocations to discipline an open-economy neoclassical growth model and demonstrate how trade and financial openness interact to drive the rate of convergence to steady state.

At present, we do not have a canonical model of bilateral investment around the world. Developing such a framework is a very promising avenue for future research. Among many potential applications, it could be used to: explain current capital allocations, assess the size of distortions (tax havens, but also trade barriers or capital controls), estimate third-country effects of shocks or policies (e.g. effect on Vietnam of a China-US specific shock), and document segmentation along geography but also political affiliation⁸ or level of development.

The data on capital allocation generally covers three areas, especially when focusing on global portfolios. The first area is data on the universe of securities outstanding at a point in time and their characteristics (e.g. for a bond: notional amount, currency, maturity, legal framework, etc.). The second area is data on who owns the securities (e.g. a mutual fund domiciled in the US, or a German bank). The third area is data on who issued the security (e.g. which firm is the immediate issuer, where is that firm registered in the world, who owns that firms, etc.). There has been remarkable progress in the last few years on all three aspects of the data. Here we focus on a brief summary comparing commercial data on holdings to official data.

The most obvious differences between commercial and official data are cost, accessibility, coverage, and level of detail. For each source of data, there are pros and cons, and we review here some of the most common that should be of interest to new researchers entering this area.

Official data are collected via a regulatory or legislative mandate. This generally provides an advantage because compliance of the reporting entities within the mandate is high. For example, in the US regulation requires that custodian intermediaries report cross-border portfolio investments to the U.S Treasury and these data are then used to compile TIC and the portfolio component of the US IIP. Coverage within the mandate is very high (e.g. all type of investors, very comprehensive within type). Similarly, the ECB SHS database is collected via regulatory disclosures. The main drawback is that the mandates often do not include important parts of the data from a research perspective. For example, the US Treasury data does not include domestic positions, so that it is not possible within that mandate to observe all the holdings of securities by an entity, but just the cross-border part. The ECB SHS data, instead, does observe the entire portfolio (domestic and foreign) of each entity, but in most cases the data made available for researcher is not at the holder level (only available at the country-sector-of holder level, i.e. all German funds together and not fund by fund).¹¹

Commercially available data comes from two sources: data collected and resold by data providers and regulatory data that is available for purchase. The main advantage of commercially available data are its availability to most researchers (direct access, often no nationality requirement) and the ability to combine various data sources to assemble the desired dataset for research. The main drawback is the potential lack of coverage of the underlying universe. In the US, for example, commercial data on holdings is available for mutual funds, ETFs, and insurance companies, but extensive coverage of pension funds is more difficult, and coverage of hedge funds is very limited. In many other countries, with the exception of Japan and Norway, securities holdings of large sovereign wealth fund are not publicly available.

As an illustrative example, consider the outward US portfolio position. The official data are from TIC. For commercial data, we use US domiciled open-end mutual funds and ETFs and insurance companies.¹² Of course, funds and insurance holdings are only a subset of total positions, and the data does not include households, banks, hedge funds, and non-financial corporations. However, the TIC annual report for 2020 shows that investment funds and insurance companies are by far the largest cross-border investor types and account for 55.34% of the United States external portfolio investment. The commercial data provides nearly universal coverage of these investor types in the US and, therefore, accounts for a large share of aggregate US foreign investment, rather than simply being a small representative slice of it. In addition, these commercial datasets contain all domestic investment positions and the data are available quarterly. Domestic positions are, of course, much larger than cross-border ones and account for the majority of these investors’ holdings.

Figure 5 compares the geographic composition of foreign equity and bond investment available in commercial data to the United States official TIC data. We compute, for each dataset, the share of overall outward investment in that asset class going to each destination. The two datasets offer a remarkably similar picture. The (across-destination) correlation among the two estimates of these shares for common equity is 0.9940 in 2007, 0.9970 in 2015, and 0.9920 in 2020, respectively. For bond investment the correlations are 0.9370, 0.9889, 0.9865, respectively. The close alignment between the data occurs for two reasons: first, the commercial data directly accounts for a large share of the total positions, and the observed positions are largely representative of total in this dimension. The correspondence between the micro-data and administrative data tightens in recent years, perhaps as a mutual funds, ETF, and insurers account for a larger share of the aggregate positions.

There certainly are dimensions along which the commercial data differs from the aggregate due to differences in investor holdings by type. An illustrative example of investor heterogeneity (even within the commercial data) is the currency composition of bond portfolios. In Figure 6, we compare the currency composition of cross-border bond portfolios held by US mutual funds and insurance companies. Even conditional on investing abroad, well over 90% of insurance company bond holdings are denominated in US dollars compared to slightly more than 60% of mutual fund foreign bond holdings. Insurers are more likely to want to hold dollar bonds to match the fact that their liabilities (e.g. insurance policies) are in dollars, while mutual funds might want to offer some foreign currency exposure to their underlying investors.

The heterogeneity between insurance and fund holdings is of more general interest with respect to the differential propensity to fire-sell assets. Coppola (2022) uses the micro-data on funds and insurers’ holdings to document (both domestically and internationally) that funds are disproportionally likely to fire sell bonds during periods of aggregate market stress. Insurers, on the contrary, are “safe hands” in a crisis, but also unconditionally trade the bonds less, making the bonds that they hold endogenously less liquid in normal times but more resilient to market crashes. Coppola (2022) cleverly uses the variation in purchases of the bonds at issuance (on the primary market) by large insurers as an instrument for holdings to provide a causal interpretation of his results.

Figure 7 shows an updated version of Figure 2 of Maggiori, Neiman, and Schreger (2020). For 9 developed countries, or currency areas in the case of the European Monetary Union, we contrast the share of domestic corporate bond investment in the investor country’s local currency (left) with the share of foreign investment in the recipient country’s local currency (right). There are two key takeaways: first, for every developed country, the overwhelming share of domestic investment is in the investor’s local currency; second, for every country except the United States, the overwhelming share of foreign investment is not in the recipient’s local currency.

In Maggiori, Neiman, and Schreger (2020), the authors show that foreign portfolio investment in corporate bonds is largely denominated either in the investors’ home currency or the international currency, the US dollar.¹⁴ Despite Figure 7 only reporting a country level aggregate, this figure shows the benefit of beginning with commercial fund data relative to standard cross-border datasets because they include domestic holdings, allowing for a clear comparison of investor heterogeneity across the domestic and foreign domain.

Using the GCAP cross-country micro-data, we next turn to documenting the heterogeneity in the way sovereigns and firms borrow internationally in the bond market. In Figure 8, we plot the share of foreign-owned sovereign and corporate debt in each country that is denominated in the issuer’s local currency. A few key features emerge. First, a wide range of both developed and emerging market sovereigns primarily borrow from foreign investors in their local currency. In an important contribution, Eichengreen and Hausmann (1999) highlighted the tendency of emerging and developing markets to borrow in foreign currency as a key source of macroeconomic fragility, terming the phenomenon “Original Sin.” Lane and Shambaugh (2010) and Bénétrix, Lane, and Shambaugh (2015) combine data from CPIS, UNCTAD, and the BIS to estimate the net currency exposure of a large panel of countries. They document significant heterogeneity between emerging and developed countries, and a major shift of developing and emerging markets away from foreign currency liabilities.

As Figure 8 makes clear, by 2020 a significant fraction of emerging markets have more than half of their foreign-held sovereign bond debt denominated in local currency while no country, emerging or developed, other than the United States raises half of their foreign-held corporate bond finance in their local currency.

There is now an extensive literature exploring the factors pushing emerging markets sovereigns to borrow in local or foreign currency. Du and Schreger (2022) document the heterogeneity between sovereign and corporate external currency compositions for emerging markets and study the interaction between the currency composition of sovereign and corporate debt as a driver of sovereign default risk. For the sovereign’s choice of the currency composition of debt, Ottonello and Perez (2019) highlight variation in the costs of inflation, Engel and Park (2022) point to an important role for commitment issues, and Du, Pflueger, and Schreger (2020) focus on the interaction between limited commitment and risk premia as a driver of the issuance decision. Wu (2020) considers how sovereign risk premia are affected by the currency composition of corporate debt. Devereux and Wu (2022) illustrate how foreign reserves can lead emerging markets to issue more sovereign debt in their own currency.Arslanalp and Tsuda (2014)a, Arslanalp and Tsuda (2014)b construct country-level datasets on the currency composition of external sovereign debt, including sectoral holdings data for many countries.

On the corporate side, Salomao and Varela (2022) consider the optimal choice of the currency composition of corporate borrowing in a dynamic model in which firms trade-off the lower cost of foreign currency debt against the volatility it induces arising from currency mismatch. Maggiori, Neiman, and Schreger (2023) provide a simple model of the currency composition of corporate debt in which debt markets are segmented by currency and firms hedge the currency risk from foreign currency debt issuance.

Corporate-Sovereign Currency Composition Puzzle. Despite the striking regularity seen in Figure 8, there is still no generally accepted explanation for the difference in the external debt composition between sovereigns and corporates. Here, we briefly discuss a few possibilities. One reason may be that sovereigns and corporates have different exposures to exchange rate fluctuations. The government’s primary source of revenue is taxes and might be heavily tilted toward local currency denominated non-tradables. By contrast, the firms borrowing in foreign currency may tilt more towards exporters, and so perhaps they differ from a risk perspective.¹⁵ Alternatively, sovereign and corporates may differ in their concern about any externalities imposed by their borrowing decisions. If foreign currency debt is a source of aggregate volatility for the economy, then perhaps the sovereign internalizes this externality but firms do not. This would lead the two types of agents to choose different compositions of the debt outstanding.¹⁶

Of course, in equilibrium foreign investors can choose which type of debt to hold out of the outstanding stock, especially as many countries have relaxed capital controls. For example, the overwhelming share of foreign-owned local currency sovereign debt in emerging markets is actually issued in their respective local market. Foreign investors that own local currency sovereign debt acquire the bonds by coming onshore. There are also significant amounts of domestically issued local currency corporate debt available for these foreign investors to buy, but the data on holdings reveal that they choose not to. We summarize this puzzling allocation in Figure 9. The pattern in the bottom left corner offers a challenge to existing theories of how countries graduate from original sin. For example, if foreign investors were previously unwilling to hold local currency bonds due to inflation risk, then once graduation occurs this should have led to foreign holdings of local-currency bonds issued by both the sovereign and corporate sector. Similarly, if foreign investors were simply unfamiliar or had a distaste for the corporates, then this should have manifested in low holdings of the foreign-currency denominated debt of the corporates.¹⁷ The micro data also excludes explanations that are purely compositional: while the set of corporates that issues in the domestic currency and in foreign currency is different, the result that foreign investors only buy the foreign-currency debt holds even within each corporate group (comparing debt issued by the same firm).

What makes foreign investors willing to hold domestic local currency sovereign debt but not domestic local currency corporate debt? Some possibilities include the desire to avoid domestic law bankruptcy procedures that are more relevant for corporate than sovereign debt, the desire of investors to separate currency and credit risk, or perhaps market frictions like differential liquidity. At present, this differential currency composition of sovereign and corporate debt remains a puzzle for models to explain. The solution might help us shed light on the key frictions in international capital markets.

If investors are not lending to corporates in the currency of the borrowing country, how are they lending? The answer is dollars or in the investors’ home currency. We begin by documenting the sense in which foreign investment is tilted to investors’ home currency. To do so, we use the GCAP data and follow Maggiori, Neiman, and Schreger (2020) and run regressions of the form:

\[ s_{j,p,c}=\alpha_{j,p}+{\beta_j}1_{{Currency_i=Country_j}}+Controls+\epsilon_{j,p,c} \tag{1}\]

where \(s_{j,p,c}\) is the share of the total holdings in our data of a particular corporate bond \(c\) issued by parent firm \(p\) that is held by investors from country \(j\), \(\alpha_{j,p}\) is a fixed effect for the parent firm, and \(1_{{Currency_i=Country_j}}\) is an indicator variable denoting that the security is denominated in the investor’s home currency. The key coefficient of interest is \(\beta_j\), measuring how much more of a given security, all else equal, investors own when it is denominated in their own currency. The key idea is that if \(\beta_j\) is zero, investors do not display any tilting of their portfolio to assets denominated in their home currency, but if it is positive, then this means that investors tilt their portfolios toward their home currencies. The firm fixed effects sharpen the focus on currency since any firm invariant characteristics (i.e. size, industry, exporter status) are absorbed.

Table 1, reports large estimates of \(\beta_j\) showing that home currency bias is remarkably strong. For example, in column 2 the estimated \(\beta_j\) for Canada is 0.849. This means that if a security is denominated in Canadian dollars, then Canadian funds own 84.9% more of the amount owned by mutual funds globally than what they hold of securities that are not denominated in Canadian dollars but issued by the same issuer. Given Canada’s small share of global portfolio investment, this is a huge degree of bias.¹⁸ In Table 3, we explore sectoral heterogeneity in home currency bias for American-domiciled mutual funds and insurance companies. While both sectors are strongly home currency biased, we showed that insurance companies are more so, echoing the findings in Figure 6.

The strength of home currency bias in explaining investors’ international bond positions naturally raises the question of the strength of this effect relative to classic home country bias. The influential work of French and Poterba (1991) found that investors disproportionately hold equity securities issued by domestic firms. The subsequent literature demonstrated that the same is true, to an even greater extent, for bonds. However, as we have emphasized, data limitations have meant that traditional analyses have not included information on currency. Maggiori, Neiman, and Schreger (2020) explore this question by adding a dummy for a security being issued by a firm from an investor’s home country to Equation 1.¹⁹ They show that conditional on the currency of denomination of a bond, the home country effect explains a surprisingly small amount of the variation in investors’ portfolios. At least for corporate bonds, inference of home-country bias is confounded by the presence of home-currency bias.

Currency trading also involves derivatives, such as currency forwards and swaps. Regulations introduced in many advanced economies in the aftermath of the Global Financial Crisis require the disclosure of derivative transactions (e.g. the European Market Infrastructure Regulation (EMIR)). The resulting data repositories offer an opportunity for researchers to better understand the nature and extent of financial hedging and currency preferences more broadly. Cenedese, Della Corte, and Wang (2021) use the EMIR data to demonstrate that balance sheet constraints cause CIP deviations at the dealer level. For mutual funds, Sialm and Zhu (2021) use new public disclosures of US-domiciled funds to document currency trading via forwards in addition to investment in foreign-currency bonds. Liao and Zhang (2020) provide evidence of hedging for pension funds and insurers and introduce a model of exchange rate determination with incomplete hedging in segmented markets. Du and Huber (2023) explore the hedging behavior of different types of investors holding US dollars around the world. Nathan and Zeev (2023) explore how hedging demand affects CIP deviations and Nathan and Zeev (2022) explore how global equity markets drive hedging demand.

In the context of home-currency bias, hedging is interesting in several dimensions. On the investor side, it is important to assess whether funds tend to hedge even those (relatively few) bonds that they do buy in foreign currency. For example, many developed sovereigns almost exclusively issue in their own currency (e.g. the US only issues in dollars) and the decision of buying that sovereign cannot be disentangled from the currency without separately trading derivatives. Second, in assessing whether the more active mutual funds use derivatives to actively trade currencies, funds may actually use derivatives to implement the carry trade rather than hedge currency risk. Sialm and Zhu (2021) provides evidence for both of these possibilities for US mutual funds.

On the firm side, an interesting question is whether the foreign currency issuance is unhedged, or whether it is operationally or financially hedged. Some larger corporations have started reporting their hedging strategy and there is evidence for some specific countries (e.g. Alfaro, Calani, and Varela (2021) for Chile), but providing systematic evidence is an open research opportunity. One potential conjecture is that financial hedging is more likely in large developed currencies that have liquid derivatives markets while operational hedging is more likely in emerging market currencies. For example, a Euro Area corporation might issue in dollars and then hedge to capture a lower cost of capital, but that same corporation issuing in Brazilian Real (BRL) might instead be more likely to do so to finance a local operation in Brazil with BRL costs or BRL-linked assets.

In Figure 11, we combine data from each country’s multilateral international investment position to illustrate the importance of tax havens as sources and destinations of cross-border investment. In panel (a) While the United States is the largest destination for investment followed by the UK, many of the next countries are tax havens such as Luxembourg, Ireland, and the Cayman Islands. Many of these destinations are recorded as receiving capital in amounts that far exceed any notion of domestic activity, so that it has long been clear to economists that the capital is just passing through to its ultimate destination.

To illustrate why so much cross-border investment is recorded as involving tax havens as destinations, it is useful to consider a concrete example: how the Brazilian oil-company Petrobras raises capital from foreign investors. Figure 12 illustrates the capital allocations for this example. One might have expected that Petrobras raised bond finance from foreign investors by selling bonds directly –i.e. as a Brazilian parent company– to foreign investors. In practice this is not the case. Crucially, however, this does not mean that Petrobras does not borrow from foreign investors. Instead, Petrobras operates a number of financing subsidiaries such as Petrobras International Finance Company (Pifco) in the Cayman Islands and Petrobras Global Finance in the Netherlands. When foreign investors purchase these bonds, they recognize that their meaningful economic exposure is to the Petrobras group. However, for the purposes of international financial statistics, a US investor owning a bond issued by Pifco is recorded as a portfolio debt investment with the Cayman Islands as destination. As we discussed in Section 2.2.1, this is because international financial statistics are recorded on a residency basis, whereby the geographic location of an activity is ascribed to the immediate counterparty (i.e. Pifco in the Caymans). Statistics that instead record economic activity according to the location of the parent or controlling entity (i.e. Petrobras parent company in Brazil) are instead referred to as nationality-based statistics.

Coppola et al. (2021) undertake a systematic effort to restate global portfolio positions on a nationality basis, with a focus on looking through the role of global tax havens. In order to do so, they match every security to both its immediate issuer and the ultimate parent (and their respective countries), and use this to restate both the stock of securities outstanding (issuance) as well as micro-data on securities holdings for large developed countries.²⁴ They estimate the share of investment different countries make in every country around the world in each asset class that is reclassified as going to another country when moving from a residency to a nationality basis. The authors focus on providing restatements of countries’ external investment positions reallocating tax haven resident investment. Abstracting from investor-specific heterogeneity conditional on investing in tax havens, Coppola et al. (2021) also provide a nationality-based restatement of global portfolio investment for all countries that report to CPIS with further breakdowns available for a subset of larger developed investors.²⁵

In Figure 13, using data from 2020, we update two findings from Coppola et al. (2021) where looking through tax havens is particularly important. In panel (a), we plot the share of 9 countries’ external bond portfolios that are allocated to the BRICS (Brazil, Russia, India, China, and South Africa). The US share increases from approximately 1% to 6%, and other countries experience comparable increases. In the case of equity positions, reported in panel (b), the largest adjustment concerns China as a destination. The share of the US external position invested in China goes from around 3% to 13%, an increase of $850 billion. As discussed in detail in Coppola et al. (2021) and Clayton et al. (2023), this massive adjustment for China arises because of the use by Chinese firms of the variable interest entity (VIE) structure. Chinese regulations forbid firms in sensitive industries, including much of the technology sector, from receiving foreign equity investment. Alibaba, Tencent, Baidu, and other Chinese tech firms side-step these restrictions with a VIE structure that simultaneously allows the firms to sell equity claims on a Cayman Island-based shell to foreign investors and tell Chinese regulators that the local companies in China only have domestic equity owners.

Nationality is not the only possible criterion that researchers can use to reallocate investments away from the residency criterion. Rather than relying on the binary classification of firms as belonging entirely to a single country, Coppola et al. (2021) also classify exposure to a country by scaling the investment position according to where a firm actually earns its revenue.²⁶ For some theoretical applications, this revenue-based measure of exposure may be more relevant than either a residency or nationality approach. Figure 14 provides an example of the difference between the three statistical notions, focusing on US portfolio investment toward China. While the residency-based US exposure to China is roughly constant as a share of investment, we see sharp increases in the nationality and sales-based measures.

Damgaard, Elkjaer, and Johannesen (2019) undertake a related exercise for FDI statistics using the IMF CDIS data. They separate reported foreign direct investment into “Real FDI” and “Phantom FDI,” where the latter indicates FDI into a shell company unlikely to undertake actual economic activity in the recorded geography. Then, using data from the OECD and Orbis, they assign Real FDI to geography of ultimate parents. There is currently no equivalent restatement of cross-border bank lending that reclassifies borrowers according to their ultimate parent.

The right panel of Figure 11 shows that tax havens also play a major role in international financial statistics as a source of capital. While the US is the largest source of cross-border investment, tax havens feature prominently as large holders of investments made all over the world. In the case of portfolio investment, much of this effect arises from the investment fund sector being headquartered in financial centers. Funds are considered the legal owners of the assets, so that, for example, all securities held by a mutual fund domiciled in Ireland are included in Irish portfolio investment assets. Ireland also reports a portfolio equity liability, corresponding to the fund shares held by foreign investors. An important question is who own the fund shares, i.e. who exactly are these funds invested on behalf of. Identifying these ultimate owners is an active research question with important implications for international macroeconomics and public finance.

Table 3 analyzes home currency bias by estimating regressions as in Equation Equation 1 for the sum of US mutual funds and insurers (“Pooled”) and then for each of the two U.S. sectors of holder separately. The first three columns use ultimate parent firm fixed effects and bond level controls (as in the main-text) and the last three columns drop the fixed effects and controls. The first three columns provide strong evidence of home currency bias across sectors using only within-firm variation. The last three columns drop the fixed effects in order to facilitate comparison of the magnitude of home currency bias across the sectors, and thus include potential compositional effects. Since the constant in these latter specifications is the estimated average share of the holdings data of non-USD denominated assets owned by each sector, we observe that U.S. mutual funds account for a much larger share of foreign currency debt than do insurance companies. By comparing the estimated \(\beta\) to the constant, we observe that insurance companies display much stronger home currency bias than do mutual funds, even though mutual funds also display very strong home currency bias themselves.